[
https://issues.apache.org/jira/browse/HBASE-30395?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
huginn updated HBASE-30395:
---------------------------
Description:
Motivation
The decompression stream fallback currently uses
BlockIOUtils.readFullyWithHeapBuffer, which allocates a temporary 1 KiB heap
buffer and copies each read into the destination ByteBuff. This adds an
avoidable heap buffer copy for on-heap ByteBuff destinations.
was:
Motivation
The decompression stream fallback currently uses
BlockIOUtils.readFullyWithHeapBuffer, which allocates a temporary 1 KiB heap
buffer and copies each read into the destination ByteBuff. This adds an
avoidable heap buffer copy for on-heap ByteBuff destinations.
Proposed change
* Add a readFully(InputStream, ByteBuff, int) helper.
* Read directly into the backing array when the destination ByteBuff is on heap.
* Preserve the existing temporary-buffer fallback for direct or non-array
ByteBuff instances.
* Update the compressed and uncompressed stream fallback paths in
HFileBlockDefaultDecodingContext.
* Add regression tests for the direct-read and fallback paths.
Compatibility
This change does not modify the HFile file format, wire format, or
configuration behavior.
Related context: HBASE-27049
Related change: https://github.com/apache/hbase/pull/8665
> Avoid heap buffer copy in decompression stream fallback
> -------------------------------------------------------
>
> Key: HBASE-30395
> URL: https://issues.apache.org/jira/browse/HBASE-30395
> Project: HBase
> Issue Type: Improvement
> Reporter: huginn
> Priority: Major
>
> Motivation
> The decompression stream fallback currently uses
> BlockIOUtils.readFullyWithHeapBuffer, which allocates a temporary 1 KiB heap
> buffer and copies each read into the destination ByteBuff. This adds an
> avoidable heap buffer copy for on-heap ByteBuff destinations.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
